FRITZ 20 and the Chess Training Equation: The Strongest Engine Is Not Always the Best Teacher

It was three in the morning in a small chess training room on Lach Tray Street, Hai Phong, and I was rewinding a tape of a practice game. No spectators. No commentators. No applause. Only the sound of wooden pieces hitting the board, the digital clock ticking down, and the sigh of a fourteen-year-old boy who had just missed a checkmate on move thirty-one. The game lasted forty-seven moves. On the scoresheet the result was a draw. On the tape it was a defeat. The boy had no idea what he had missed until I froze the frame and showed him. On move thirty-one his clock still had four minutes and twelve seconds left. One knight move to the centre and the game was over. What made me stop on that frame was not the board. It was the laptop sitting beside it. On screen, a chess engine was running an analysis mode, and the boy's thirty-first move had been flagged with two question marks. The machine knew first. The boy knew later. The gap between those two moments is the whole story of modern chess training. That is also why I sat down to write about FRITZ 20, a chess program that markets itself with three very human titles: personal trainer, toughest opponent, strongest ally. Those three titles are a promise. And like every promise in sport, it is only worth something when it survives cross-checking. The Fritz line began in the early 1990s, tied to two names rarely mentioned in Vietnamese sports pages: Frans Morsch and Mathias Feist. In 2026, at the Intel Grand Prix in London, the version known as Fritz 3 beat Garry Kasparov in a serious game. It was the first time a commercial chess program had defeated the reigning world champion. Seven years later, at Brains in Bahrain in 2026, Deep Fritz drew 4-4 with Vladimir Kramnik. Four years after that, in Bonn in 2026, Deep Fritz beat Kramnik 4-2. The man-versus-machine race was only the glossy outer layer. Beneath it lies another story, quieter and more important. Once engines had decisively overtaken humans, their value shifted: from performing opponent to training instrument. FRITZ 20 belongs to that second era. To understand the second era you have to look at the Elo scale. A serious beginner usually sits somewhere between 800 and 1200. An international master hovers around 2400. The world's leading players occupy the band between 2700 and 2830, with the historical peaks belonging to Garry Kasparov at 2851 and Magnus Carlsen at 2882. The strongest chess engines today, when measured on machine rating lists, pass 3600. On this scale, a hundred-point gap corresponds to an expected score of roughly sixty-four percent. The distance between the world's number one and a top engine therefore approaches eight hundred points, wider than the distance between an international master and someone who has just learned the rules. Nobody learns anything from an opponent eight hundred points above them. And nobody learns anything from an opponent eight hundred points below them. The problem facing every modern chess training program is not making the engine stronger. It is teaching the engine to lower itself to exactly the distance at which the student can still learn. In Vietnam this story has its own backdrop. Le Quang Liem and Nguyen Ngoc Truong Son have both been ranked among the world's top one hundred players. Pham Le Thao Nguyen, Hoang Thanh Trang and Vo Thi Kim Phung have carried Vietnamese women's chess onto the international stage. Behind that elite group stand tens of thousands of amateur players, most of them self-taught, most without a coach, and most of whom began playing online during the pandemic years. I rewind the SEA Games 29 footage to find what people left behind. This time I was rewinding a chess game. And what gets left behind in most training sessions I have watched is not opening knowledge. It is the measurement of learning itself. FRITZ 20 advertises three functions. I test each one with the only question a documentary maker must always ask: where does it change the learner's behaviour? The role of the teacher sits in adaptive practice play. Instead of placing the student against an all-powerful machine, the software adjusts its own strength to the opponent's level, and keeps adjusting inside a single game. In other words, the engine deliberately ties its own hands to keep the game inside the zone where the student can still understand what is happening. This is the most interesting part methodologically. In educational psychology, the effective learning zone is the one where the task is just hard enough, neither so easy that it bores nor so hard that it paralyses. In chess, that zone is measurable through expected score. Winning one hundred percent of your practice games is a waste of time. Losing one hundred percent is also a waste of time. The learning window sits around a score of fifty to sixty percent. An engine that adjusts its own strength has higher training value than the strongest engine, because it solves the hardest problem in self-study: keeping difficulty inside the learning window. The role of the opponent sits in raw strength, and here lies a paradox beginners rarely notice. The strongest engine is not the most human one. A top engine chooses moves no player would consider, sometimes for reasons the human eye cannot see in ten minutes. Training against it daily can build a harmful habit: the student becomes used to defending against blows no human opponent would ever throw, while forgetting the blows humans throw all the time. The value of a training opponent therefore does not lie in the highest Elo. It lies in the ability to simulate human patterns of error. An engine that blunders at the right moment and misses a tactic in the right place is more useful than an engine that never errs. The role of the ally is the driest part and also the decisive one. The engine sits beside the student after the game, pointing out which move lost the advantage, which move missed a chance, and why. In endgames this work has reached absolute precision: modern endgame tablebases solve seven-piece positions perfectly, with no guesswork. In openings, the software manages the user's repertoire, reminding them which lines they have prepared and which they are forgetting. I trust footage more than testimony, because footage cannot lie. An analysis engine is the same, provided the user reads the whole output instead of looking only at the first move it suggests. This is where I see the generational shift most clearly. In 2026, when I started working in chess, looking up an old game meant digging through paper magazines or waiting for someone to bring a floppy disk. Today most of chess history sits one click away, and a fourteen-year-old in Hai Phong holds more data than a national team coach did thirty years ago. But more data does not automatically produce better players. Every documentary is a running track: the audience sees the finish line, I live at every starting block. Every chess training session is the same: outsiders see only the result, while those inside live through every move, every hesitation before touching a piece. Based on my experience watching matches and youth squad training sessions, a player between fourteen and eighteen misses on average four to six clear tactical chances in a thirty-minute rapid game. What matters is that most of those chances are never mentioned again in the next session. The mistake is discovered, then forgotten. An effective training routine has three phases: play, analyse, record. The recording phase is the most neglected. Serious players do not need a stronger engine. They need a system that stores what they have learned, so the same mistake does not return in the fourth month. The structure of a modern chess training programme therefore resembles the three-chapter structure I once built for a documentary about a Vietnamese female player: Doubt, Proof, Inspiration. The Doubt chapter is when the student re-examines everything they believe, including openings they have played for ten years. The Proof chapter is when the engine supplies evidence, and the student must endure the discomfort of being contradicted. The Inspiration chapter is when the lesson leaves the screen and walks into a real game, with a real opponent and a real clock. The difference between a good training program and an outstanding one lies in the Inspiration chapter. An engine can prove the student was wrong. It only becomes useful when the student still wants to sit down after learning they were wrong. Here I have to say something I have kept for many years. In 2026, when I entered the chess world as a player and then a tournament organiser, the sentence I heard most often was not a technical assessment. It was a variant of what a coach once told Quach Thi Lan: girls should not run hurdles. In chess, the version of that sentence is: what would a girl know about chess. I have run hurdles all my life, I just was never timed. In chess the measurement is very simple: the clock, the scoresheet, the Elo rating. There is no room for remarks about voice, posture or appearance. A chess engine does not know whether the person sitting in front of it is male or female. It only knows which move is best. That neutrality, technically speaking, is the greatest advantage a training tool can offer a young female player. Having said that, I am obliged to contradict myself. The strongest engine is not the best teacher, and personalised training cannot be handed over entirely to software. A human coach does three things no engine can replace. They know when a student is exhausted and needs rest, when to push, and when a defeat should be described more gently than reality. They know whether a person's pathway is three years or ten, not one game. And they know there are periods when a player improves technically while sliding backwards mentally. Top engines have also produced a flattening of style. When everyone analyses with the strongest machines and plays the moves the machine rates best, tactical diversity in junior events shrinks. Watching a game between two young players today, spectators often see two sides playing so similarly that they are hard to tell apart. That is the rarely mentioned price of a training culture built entirely on engines. The claim of a personal trainer therefore needs careful reading. What the software provides is infinite repetition and instant, tireless feedback. Its value lies in the most boring part of learning: rechecking, cross-referencing, classifying, archiving. Precisely because it handles the boring part well, the human coach gains time for the part that cannot be mechanised: reading psychology, planning a career, and teaching how to endure pressure at the two-hundredth minute of a long game. At this point I must go back and check my own data. I wrote earlier that the engine is a gender-neutral tool. Technically true. But set beside data on access, that neutrality is far narrower than it appears. A training program is only useful to someone with a strong enough computer, three hours a day, and a quiet space to sit in. Whether young female players in many localities have all three, I do not have the data to conclude, and I refuse to make a claim without evidence. But I have enough evidence to say that money still flows into publicity more than into long-term training scholarships. A women's tournament staged to mark corporate social responsibility does not produce a single extra female master. A three-year scholarship, with a coach, with software, with a computer, might. The difference between those two ways of spending lies here: one needs a photograph. The other needs a spending tracker running for thirty-six months. There is one more question buyers of the software should answer for themselves. When the strongest open-source engines in the world are released free of charge, why is there still room for a commercial program? The answer lies in the difference between an engine and a workflow. A free engine gives the user computing power. It does not give them a training system, does not manage an opening repertoire, does not track progress across six months, and does not remind them that last month they made this exact mistake twice. People who buy training software are not paying for Elo. They are paying for the organisation of work. To see why organisation matters so much, look at speed. Modern engines calculate tens of millions of positions per second on a well-configured personal computer, thousands of times faster than a grandmaster can analyse in the same span. The branching factor of chess is around thirty-five, meaning any position contains roughly thirty-five legal moves, and that figure multiplies exponentially with every move. Against that gap, the engine being right almost all the time is no longer surprising. What matters is how the learner handles that correctness. The pandemic pushed online chess into millions of Vietnamese homes. Playing platforms recorded major growth in new accounts, and chess became one of the sports least affected when venues closed. But a paradox came with it: the more chess people played online, the more the average quality of their analysis fell, because most players press the button for a new game rather than the button to review the last one. If I were shooting a film about a modern chess training session, I would place cameras in three positions: behind the student to see the screen, beside the real board to see the hands, and opposite to see the face. Those three angles correspond to three kinds of data: the move, the time, and the emotion. The engine reads only the first. That is its limit, and it is also why people still need people. Six months after that frame in Lach Tray, I returned to the training room. The boy had gained about one hundred and twenty Elo points. There was no miracle. He had simply started writing down every mistake the engine flagged, once a week, in a notebook. By the fourth month he realised he had made the same kind of error on the kingside in seven different games. That was when the learning genuinely began. The next generation of Vietnamese chess players will not grow up short of material. They will grow up with too much data. Their problem will not be finding a strong enough engine, but living alongside an engine that is always right. When a tool becomes too good, the question shifts from the tool to the user. What will distinguish an eighteen-year-old player in 2030 from thousands of peers with the same engine, the same database, the same online courses? Perhaps only one thing that cannot be downloaded: the ability to sit still with a difficult position for forty minutes without opening the machine. And perhaps that is the most important thing to say about FRITZ 20. It is most useful not when it is strongest, but when the user knows when to switch it off.